Thirty-six years of HDB resale prices, in cash and in today’s money

The HDB resale register is public back to January 1990, which makes it one of the longest continuous records of housing transactions anywhere. Over 146 complete quarters, 1990 Q1 to 2026 Q2, the median resale price per square foot has risen to 9.3 times its 1990 level in cash terms, and 4.8 times in today’s money. Those two numbers are the article. The first is the one that gets quoted; the second is the one that tells you what happened to a household.

Why publish both series?

A price in cash terms answers “what would this have cost then”. A price in today’s money answers “how much of a household’s spending power did it take”. Over three decades those diverge a long way, because the currency itself moved. Restating the whole series against the Consumer Price Index is the only way to compare 1990 with today without the comparison being mostly about inflation.

Almost every published account of this market is cash-only. That is not dishonest, but it systematically overstates how much dearer housing has become, and it makes flat decades look like rising ones.

What do the two series show?

The first complete quarter records S$65.2 per square foot in cash terms, which is S$126 in today’s money. Each row below is the last complete quarter of its year.

What does the cash series hide?

Read the two rate columns side by side and the shape of the market changes. There are stretches where the cash figure drifts upward while the real figure falls, which is a market losing ground quietly. There are also periods where both rise sharply, and those are the genuine booms.

The longest of those stretches ran from 1997 Q2 to 2010 Q2: 13 years during which the rate in today’s money did not get back to where it had already been. It was not a plateau. The rate reached S$480 per square foot in 1997 Q2, fell to S$336 by 2002 Q3, a drop of 30% in real terms, and took until 2010 to recover. In cash terms much of that period looks like a market going sideways or gently up, which is the same data telling a different story.

This is the single most useful thing an inflation correction does. It separates “prices went up” from “housing got more expensive”, and those are different claims that the same series can support or refuse depending on how it is stated.

How the series is built

Each point is the median price per square foot of every resale registered in that quarter, with a 95% interval from the order statistic. There is no statistical outlier fence on this register: what is applied is an impossibility band, and it currently withholds nothing at all, because the register is clean. Floor area comes from HDB’s own record on the same row as the price, so unlike a matched dataset there is no address join and no match rate to report.

Today’s-money figures are the cash figure divided by the Consumer Price Index for that month and multiplied by the index at the anchor. The part quarter at the end of the series is excluded from everything above: a quarter with one month in it is not a quarterly observation, and treating it as one is how a market gets reported as turning when it has not.

What it does not tell you

Three things, and they matter before this series is used to argue anything. It is a median over whatever sold, so a quarter in which larger or better-located flats happened to change hands will read higher for that reason alone; the mix-adjusted rate on each area page exists to address that and is the better instrument for a short comparison.

It is national, and this is a market where Queenstown and Woodlands are different markets rather than different prices in one. And it says nothing about the lease, which in Singapore is a price variable rather than a legal footnote: a flat with thirty years left and one with ninety sit in the same median and are not the same asset.

Figures on this page are read live from the published dataset and are current to August 2026. The quarterly series behind the table, including the confidence interval on every point, is in the same file, and the dataset page says how to take it.